NVIDIA’s New AI Servers Run On Hotub Coolant And Don’t Need Evaporators - Hackaday

NVIDIA's latest AI servers introduce an innovative liquid cooling approach using hot coolant instead of traditional evaporative cooling, significantly reducing water consumption. Based on the Ruben ar... Read original →

NVIDIA’s shift to hot-liquid cooling isn’t just about efficiency—it redefines thermal management as a first-class architectural component. This move pressures upstream cold plate and pump suppliers to adapt to the extreme heat flux of 3nm nodes, while forcing hyperscalers to rebalance PUE against water usage effectiveness (WUE). With EU energy directives and California water regulations tightening, evaporative cooling is becoming a compliance liability; NVIDIA’s 45°C loop sidesteps this by enabling dry heat rejection. AMD lacks full-stack integration to respond quickly, and Intel’s delayed process nodes hinder co-design. Within 18 months, liquid cooling will be table stakes for AI servers—but the real value will accrue to materials firms mastering microchannel cold plates and high-temp dielectric fluids, not chip designers.